Migrating from Plugin SDKv2 to the Plugin Framework
The Plugin Framework is required for net-new resources and data sources;
SDKv2 is maintenance-only. Migration is per-resource and incremental: a
muxed provider serves SDKv2 and Framework implementations side by side, so
you never need a big-bang rewrite. This skill covers the mux setup, the
per-resource workflow, and the behavioral traps that turn a mechanical
translation into a silent breaking change.
Reference (load when needed):
references/schema-mapping.md
— the full SDKv2 → Framework translation
table with code pairs
Decide Whether to Migrate at All
Migration has real risk and little user-visible payoff, so triage first:
- Do not migrate complex or heavily-used resources without a driving
need (a Framework-only feature, a bug that SDKv2 cannot fix). The two
SDKs differ behaviorally — most importantly around null versus zero
values — and those differences surface as breaking changes for existing
users. This is the standing policy in large providers like
terraform-provider-aws.
- Simple resources migrate safely: flat schemas, no ,
no , no , no complex nested blocks.
- New capabilities never require migrating old code — mux and write the new
resource in the Framework alongside the old ones.
To tell what mode a provider is in, check
:
present means it already serves both; only
means
SDKv2-only (mux setup is your first step); only
terraform-plugin-framework
means the migration is done.
Step 1: Mux the Provider
Combine both plugin servers in
. Serving protocol version 6
requires upgrading the SDKv2 server with
(protocol 6 needs
Terraform CLI >= 1.0; if you must support 0.12+, mux at protocol 5 with
/
instead — but the Framework provider then
cannot use protocol-6-only features like nested attributes):
go
package main
import (
"context"
"flag"
"log"
"github.com/hashicorp/terraform-plugin-framework/providerserver"
"github.com/hashicorp/terraform-plugin-go/tfprotov6"
"github.com/hashicorp/terraform-plugin-go/tfprotov6/tf6server"
"github.com/hashicorp/terraform-plugin-mux/tf5to6server"
"github.com/hashicorp/terraform-plugin-mux/tf6muxserver"
"example.org/terraform-provider-examplecloud/internal/provider"
sdkprovider "example.org/terraform-provider-examplecloud/internal/sdkprovider"
)
func main() {
var debug bool
flag.BoolVar(&debug, "debug", false, "run with support for debuggers")
flag.Parse()
ctx := context.Background()
upgradedSDKServer, err := tf5to6server.UpgradeServer(
ctx,
sdkprovider.Provider().GRPCProvider,
)
if err != nil {
log.Fatal(err)
}
providers := []func() tfprotov6.ProviderServer{
providerserver.NewProtocol6(provider.New(version)()),
func() tfprotov6.ProviderServer { return upgradedSDKServer },
}
muxServer, err := tf6muxserver.NewMuxServer(ctx, providers...)
if err != nil {
log.Fatal(err)
}
var serveOpts []tf6server.ServeOpt
if debug {
serveOpts = append(serveOpts, tf6server.WithManagedDebug())
}
err = tf6server.Serve("registry.terraform.io/example/examplecloud",
muxServer.ProviderServer, serveOpts...)
if err != nil {
log.Fatal(err)
}
}
Mux requirements that bite in practice:
- Provider schemas must match exactly across both plugins — same
provider-level attributes, same types, same descriptions. Keep one source
of truth for the provider configuration and mirror it.
- Each resource and data source may exist in only one of the two
plugins. Migration's final step is deleting the SDKv2 registration.
- If publishing to the Registry with protocol 6, set
"metadata": {"protocol_versions": ["6.0"]}
in
terraform-registry-manifest.json
.
Step 2: Baseline Before You Touch Anything
The migrated resource must be indistinguishable to users. Prove it with
tests that exist before the migration:
- Ensure the resource has passing acceptance coverage: with an
import step (), , and per-attribute
update tests. If coverage is missing, write it against the SDKv2
implementation first — these tests are the migration's acceptance
criteria and must pass unchanged afterward.
- Note behaviors tests don't capture: attribute defaults, what happens
when optional attributes are omitted (null vs // is about
to matter), and any / normalization.
Step 3: Port the Resource
Translate schema and CRUD using the mapping table in
references/schema-mapping.md
. The rules that prevent breaking changes:
- Blocks stay blocks. An SDKv2
Elem: &schema.Resource{...}
written as
syntax in user configs must become a Framework Block
(/) — converting it to a nested
attribute changes the HCL syntax users must write, which is a breaking
change. Nested attributes are for new schema only.
- Null is not zero. SDKv2 returned for unset;
the Framework model gives you that distinguishes null,
unknown, and . Everywhere the old code checked or relied on
, decide explicitly what null means, and make sure you send the
API the same thing SDKv2 sent (usually: omit the field when null).
- Keep the attribute. Net-new Framework resources may omit a
redundant , but a migrated resource must keep its exact schema —
removing or renaming attributes breaks existing state and configs.
- State must round-trip. The Framework reads the state SDKv2 wrote. If
every attribute keeps its name and type, no state upgrade is needed. If
the old schema stored a value the new types package normalizes
differently, you need a — treat that as a signal the
resource may be in the do-not-migrate bucket.
Step 4: Move the Registration
Register the resource in the Framework provider's
and delete
it from the SDKv2 provider's
in the same commit — mux errors
on duplicates.
Step 5: Verify
- The pre-existing acceptance tests pass without modification —
especially , which diffs imported state against
stored state and catches most null-vs-zero regressions.
- Add a state-compatibility step: apply a config with the last released
(SDKv2) provider version, then plan with the migrated build — the plan
must be empty. In this is a two-step test
using for the old version, then
with asserting an empty
plan. The skill (if available) documents the
pattern.
- against a real pre-migration state file shows no diff.
Checklist
Related Skills
Use the
skill (if available) for Framework CRUD,
finder, and waiter patterns in the ported code, and
for the regression and version-upgrade test patterns.